資料挖掘(Data Mining)運用在衛星影像並結合機器學習理論,可從大量資料中發現知識。採用迴歸樹(CART)法獲得水稻分類知識是應用範例之一雖然此方法仍需進行樣本選取,但可降低樣本選取因素對分類成果的影響,如範圍狹小,特性不均勻等因素。利用機器學習集成方式Boosting,可以解決樣本重新選取所造成的特性改變,有助於提升分類成果準確度。利用Boosting組合分類準確度較差之樣本,分析並獲得分類精度較佳的樣本組合,避免重新取樣的問題,再配合CART分類法的使用,可提昇分類準確度。研究成果顯示,在分類精度方面利用Boosting方法比傳統最大概似法及CART法,分別提昇了近5%及3%。
Data Mining can be applied to the satellite images and can be combined with machine learning theory. This technology is used to discover knowledge from large mounts of data. CART is a kind of methods to acquire the knowledge of rice paddy classification. Although this method needs to select the samples, it can reduce the effects which are caused by the selection of samples on the results of classification, for example, the narrow area and the uneven characteristics. The machine learning method, Boosting, can solve the problem of characteristic changes that are caused by the re-selection of samples. This method can increase the accuracy of the classification. The samples which have the low accuracy of classification are organized by Boosting Method. Boosting Method analyzes the samples and acquires the samples that have the higher classification accuracy in order to avoid the re-selection of samples. And, Boosting method co-operates with CART classification method to improve the accuracy of classification. According to the result of this study, Boosting Method can improve the maximum likelihood method and the CART method on the classification accuracy which rise 5% and 3% individually.